PREDICTIVE POLICING: THE ARGUMENT FOR PUBLIC TRANSPARENCY ERIK BAKKE* Introduction ................................................ 132 I. Predictive Policing ................................... 134 A. Summary of Predictive Policing Technology ..... 134 B. Benefits of Predictive Policing ................... 137 C. Hazards of Predictive Policing ................... 139 D. How Law Enforcement Obtains the Technology . 140 II. Transparency: The Current Arguments .............. 142 A. Routes to Information ........................... 142 B. Arguments for Transparency..................... 144 III. Additional Arguments for Transparency Specific to Predictive Policing ................................... 147 A. Transparency as an Additional Deterrent ........ 147 1. Improved Effectiveness through Broader Trust in Law Enforcement ................... 147 2. Improved Effectiveness through a Deterrence Effect ........................................ 149 B. Transparency as an Accountability Gap-Filler .... 151 C. Transparency as Guarantee of Fairness .......... 153 IV. Restrictions on Public Access to Predictive Policing Information ......................................... 155 A. Open Records Law .............................. 155 B. The Business Information Exception............. 158 1. The Structure and Applications of the Exception ................................... 158 2. The Rationale for Extending the Business Information Exception to Predictive Policing ..................................... 160 C. The Law Enforcement Exception ................ 163 1. The Exception in General ................... 163 2. The Exception’s Applicability to Predictive Policing ..................................... 165 3. Reasonably Segregable Exception ............ 168 * J.D. 2018, New York University School of Law. Special thanks to Professors Barry Friedman and Maria Ponomarenko, as well as the Policing Project. Additional thanks to Natalie C. Gow and the entire editorial staff at the Annual Survey of American Law for all of their input and efforts preparing this Note for publication. 131 132 NYU ANNUAL SURVEY OF AMERICAN LAW [Vol. 74:131 D. The Unliklihood of Accessing Predictive Policing Information ..................................... 169 Conclusion ................................................. 171 INTRODUCTION At 4 p.m. on a Tuesday in Santa Cruz, the officer’s squad car radio reports a robbery. The officer puts his car in gear and radios back his pursuit of the robber.1 After making a few turns, he catches up to the suspect running from the scene just moments af- ter receiving the initial report. The officer’s success is not due to luck, but to the Santa Cruz Police Department’s (SCPD) predictive policing system. Santa Cruz receives a significant amount of tour- ism, and with that tourism comes a substantial amount of street crime. Fast response times are crucial for tracking down suspects. Looking to improve response times, the SCPD contracted with PredPol, a private company, to purchase predictive policing technology. PredPol’s technology, like competing predictive policing tools, uses computing and analytics to make predictions about where and when crimes are most likely to occur. The PredPol system estimates the likelihood of four crimes (auto theft, vehicle burglary, burglary, and gang-related activity) each day for fifteen zones. Each zone cov- ers a precise area of only 500 by 500 feet. The SCPD uses those predictions to determine where to station its officers. Police departments using predictive technologies hope they will improve their departments’ effectiveness. PredPol allows the SCPD to pursue that goal without any additional hires, which its budget does not allow. The system is not the only new measure Santa Cruz is implementing to try to improve policing. The SCPD has also instructed officers to engage more with the community and forge more relationships with locals as they patrol, but PredPol re- mains a substantial part of the city’s plans for improving police tactics. PredPol is just one of many predictive policing algorithms available for purchase, and Santa Cruz is just one of many police departments to have adopted the technology. Predictive policing devices and technologies are growing in popularity among law en- forcement departments, with thirty-eight percent of police depart- 1. The introductory story and application in Santa Cruz come from an article by Eyragon Eidam, The Role of Data Analytics in Predictive Policing, GOV’T TECH. (Sep. 2016), http://www.govtech.com/data/role-of-data-analytics-in-predictive-policing .html. 2018] PREDICTIVE POLICING 133 ments currently using predictive policing and seventy percent expecting to implement it in the next two to five years.2 But as their popularity is increasing, so too are concerns about the potential downsides of the technology. Law enforcement and the public are not entirely aware of every hazard that will accompany predictive policing technologies. Due to the highly technical nature of predic- tive policing, public transparency should be a priority. While real- time predictions of crime locations must be withheld for the tech- nology to provide any real benefit, police should at the very least reveal algorithm inputs, algorithms, and obsolete predictions when- ever possible if employing predictive policing. In Section I, I describe what predictive policing is, its benefits, and its drawbacks. I also explain how law enforcement typically purchases the technology from private developers and why private development raises concerns about information asymmetry. In Section II, I describe the most common arguments for and against predictive policing. In Section III, I add my own arguments. First, arguing the importance of transparency specific to predictive policing, I contend transparency provides a critical democratic check for predictive policing. Second, I argue that transparency can increase law enforcement effectiveness in jurisdictions with predic- tive policing through increased legitimacy and deterrence. Finally, I propose that transparency allows the public to demand change in cases where flawed algorithms arbitrarily distribute the costs of en- forcing the law. In Section IV, I describe the current apparatus through which the public can request transparency from the government. Open records laws, at the federal and state level, require the government to provide answers upon request from citizens, but they also con- tain a number of exceptions. Here, the business information and law enforcement exceptions both provide potential avenues for de- nying public information requests. 2. OFFICE OF CMTY. ORIENTED POLICING SERVS., U.S. DEP’TOF JUSTICE, FUTURE TRENDS IN POLICING (2014).As a whole, the predictive analytics industry is expected to grow from USD $ 3.49 billion to $10.95 billion by 2022. Zion Market Research, Trends in Predictive Analytics Market Size & Share will Reach $10.95 Billion by 2022, GLOBENEWSWIRE (Mar. 2, 2018), https://globenewswire.com/news-release/2018/ 03/02/1414176/0/en/Trends-in-Predictive-Analytics-Market-Size-Share-will- Reach-10-95-Billion-by-2022.html [https://perma.cc/99RP-LMJC]. Over 60 U.S. police departments currently use some form of predictive policing. ANDREW GUTH- RIE FERGUSON, THE RISE OF BIG DATA POLICING: SURVEILLANCE, RACE, AND THE FU- TURE OF LAW ENFORCEMENT(2018). 134 NYU ANNUAL SURVEY OF AMERICAN LAW [Vol. 74:131 I. PREDICTIVE POLICING Predictive policing is the application of analytical techniques to identify where crime is likely to occur and who is likely to commit crime.3 It refers to both traditional methods of distributing law en- forcement resources where departments believe they will most be needed and modern predictive algorithms and codes, the latter of which are the focus of this article.4 This section defines the modern high-tech approach to policing, weighs its benefits against its risks, and describes how law enforcement usually obtains predictive polic- ing technology. A. Summary of Predictive Policing Technology Predictive analytics rely on “sophisticated computer programs, extensive data sets, and professional data analysts.”5 Advanced tech- nology helps to “uncover non-obvious relationships” and make pre- dictions that were previously impossible.6 One of predictive policing’s most distinctive characteristics is its ability to predict fu- ture behavior from past data, moving beyond the dataset of past crimes.7 Consider a town with one hundred convenience stores. Old predictive methods could use previous robberies at three con- venience stores to predict future robberies at those stores or stores near them. A new predictive policing algorithm might combine past crimes with other data to inform officers of a greater chance of a robbery occurring at a fourth convenience store across town. The technology provides a more granular prediction, and algorithms al- 3. See id. at 1 (defining predictive policing). In 2009, the National Institute of Justice defined it as “taking data from disparate sources, analyzing them, and then using the results to anticipate, prevent and respond more effectively to future crime.” Beth Pearsall, Predictive Policing: The Future of Law Enforcement?, 266 NIJ J. 16, 16 (2014). 4. See generally, WALTER L. PERRY ET AL., PREDICTIVE POLICING 9 (RAND Corpo- ration 2013). 5. Tal Z. Zarsky, Transparent Predictions, 2013 U. ILL. L. REV. 1503, 1510 (2013). 6. PERRY, ET AL., supra note 4, at 2; See also, Andrew G. Ferguson, Predictive Policing and Reasonable Suspicion, 62 EMORY L.J. 259, 281–84